CVE-2026-58116
CriticalCVSS 9.8Exploitation Probability (EPSS)
Elevated risk60th percentile - higher than 60% of all known CVEs
Summary
LLaMA-Factory through version 0.9.5 contains a remote code execution vulnerability that allows attackers with WebUI access to execute arbitrary Python code by supplying a malicious model path in the Chat or Training interfaces. The application passes user-supplied model path input unvalidated into AutoTokenizer.from_pretrained() and AutoModel.from_pretrained() with a hardcoded trust_remote_code=True parameter, causing the Hugging Face transformers library to fetch and execute arbitrary code from a remote or local model repository with the privileges of the server process.
Risk Assessment
An attacker can take over the server, steal data, or disrupt service by leveraging the full privileges of the server process.
Recommendation
Immediately update LLaMA-Factory to a version newer than 0.9.5 and restrict WebUI access to trusted users only.
Other vulnerabilities in LLaMA-Factory
See all- CVE-2026-85673High
LLaMA-Factory contains a server-side request forgery vulnerability in the OpenAI-compatible API multimodal media URL handler that allows unauthenticated attackers to bypass SSRF validation. The check_ssrf_url guard validates URLs once but requests.get follows redirects and re-resolves DNS without re-validation, enabling attackers to use HTTP redirects or DNS rebinding to access internal addresses and cloud metadata endpoints.
- CVE-2025-61784High
LLaMA-Factory before version 0.9.4 contains an SSRF vulnerability in the chat API that allows any authenticated user to force the server to make arbitrary HTTP requests to internal and external networks. The same mechanism also allows for Local File Inclusion (LFI), enabling reading arbitrary files from the server's filesystem. The issue is in the `_process_request` function in `src/llamafactory/api/chat.py`, which does not validate URLs.
Original NVD description (English source)
LLaMA-Factory through 0.9.5 contains a remote code execution vulnerability that allows attackers with WebUI access to execute arbitrary Python code by supplying a malicious model path in the Chat or Training interfaces. The application passes user-supplied model path input unvalidated into AutoTokenizer.from_pretrained() and AutoModel.from_pretrained() with a hardcoded trust_remote_code=True parameter, causing the Hugging Face transformers library to fetch and execute arbitrary code from a remote or local model repository with the privileges of the server process.
Vulnerability data from NVD (NIST) · CISA KEV · EPSS